The pathologic grade of endometrial adenocarcinoma is widely recognized as an important prognostic and therapeutic indicator. However, disagreement persists about the optimal method of determining grade. The pathology committee of the Gynecologic Oncology Group employs a system based on the proportion of tumor in glandular array; this system is both reproducible and predictive of outcome. Others have suggested that grading based on nuclear pleomorphism and the size of nucleoli provides better prognostication. We compared the three-level architectural grading system (AG) with a two-level nuclear grading system (NG) to determine reproducibility and prognostic value in 88 cases of stage 1 endometrial adenocarcinoma. Three pathologists made independent assessments of grade by each method. The division of tumors by architectural arrangement was superior for predicting survival (83%, 73%, and 44%, AG, p = 0.005; vs. 79% and 61%, NG, p = 0.14) and equivalent to nuclear grading for prediction of recurrence. Both systems were moderately reproducible (k = 0.49, AG; k = 0.57, NG). Assessment of NG was more tedious than that of AG. Subdivision of architectural grade based on high nuclear atypia, as recommended in current, International Federation of Gynecology and Obstetrics guidelines, did not improve prognostication. Because grading based on nuclear pleomorphism does not provide prognostic information superior to that resulting from architectural grading, we do not advocate its use in routine surgical pathology practice.